1 /* -*- MODE: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
5 * Lustre Metadata Server (mdd) routines
7 * Copyright (C) 2006 Cluster File Systems, Inc.
8 * Author: Wang Di <wangdi@clusterfs.com>
10 * This file is part of the Lustre file system, http://www.lustre.org
11 * Lustre is a trademark of Cluster File Systems, Inc.
13 * You may have signed or agreed to another license before downloading
14 * this software. If so, you are bound by the terms and conditions
15 * of that agreement, and the following does not apply to you. See the
16 * LICENSE file included with this distribution for more information.
18 * If you did not agree to a different license, then this copy of Lustre
19 * is open source software; you can redistribute it and/or modify it
20 * under the terms of version 2 of the GNU General Public License as
21 * published by the Free Software Foundation.
23 * In either case, Lustre is distributed in the hope that it will be
24 * useful, but WITHOUT ANY WARRANTY; without even the implied warranty
25 * of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
26 * license text for more details.
29 # define EXPORT_SYMTAB
31 #define DEBUG_SUBSYSTEM S_MDS
33 #include <linux/module.h>
34 #include <linux/jbd.h>
36 #include <obd_class.h>
37 #include <lustre_ver.h>
38 #include <obd_support.h>
39 #include <lprocfs_status.h>
40 /* fid_be_cpu(), fid_cpu_to_be(). */
41 #include <lustre_fid.h>
43 #include <linux/ldiskfs_fs.h>
44 #include <lustre_mds.h>
45 #include <lustre/lustre_idl.h>
47 #include "mdd_internal.h"
49 static struct lu_object_operations mdd_lu_obj_ops;
51 static inline void mdd_set_dead_obj(struct mdd_object *obj)
54 obj->mod_flags |= DEAD_OBJ;
57 int mdd_la_get(const struct lu_env *env, struct mdd_object *obj,
58 struct lu_attr *la, struct lustre_capa *capa)
60 struct dt_object *next = mdd_object_child(obj);
61 LASSERT(lu_object_exists(mdd2lu_obj(obj)));
62 return next->do_ops->do_attr_get(env, next, la, capa);
65 static void mdd_flags_xlate(struct mdd_object *obj, __u32 flags)
67 obj->mod_flags &= ~(APPEND_OBJ|IMMUTE_OBJ);
69 if (flags & LUSTRE_APPEND_FL)
70 obj->mod_flags |= APPEND_OBJ;
72 if (flags & LUSTRE_IMMUTABLE_FL)
73 obj->mod_flags |= IMMUTE_OBJ;
76 struct lu_buf *mdd_buf_get(const struct lu_env *env, void *area, ssize_t len)
80 buf = &mdd_env_info(env)->mti_buf;
86 const struct lu_buf *mdd_buf_get_const(const struct lu_env *env,
87 const void *area, ssize_t len)
91 buf = &mdd_env_info(env)->mti_buf;
92 buf->lb_buf = (void *)area;
97 struct mdd_thread_info *mdd_env_info(const struct lu_env *env)
99 struct mdd_thread_info *info;
101 info = lu_context_key_get(&env->le_ctx, &mdd_thread_key);
102 LASSERT(info != NULL);
106 struct lu_object *mdd_object_alloc(const struct lu_env *env,
107 const struct lu_object_header *hdr,
110 struct mdd_object *mdd_obj;
112 OBD_ALLOC_PTR(mdd_obj);
113 if (mdd_obj != NULL) {
116 o = mdd2lu_obj(mdd_obj);
117 lu_object_init(o, NULL, d);
118 mdd_obj->mod_obj.mo_ops = &mdd_obj_ops;
119 mdd_obj->mod_obj.mo_dir_ops = &mdd_dir_ops;
120 mdd_obj->mod_count = 0;
121 o->lo_ops = &mdd_lu_obj_ops;
128 static int mdd_object_init(const struct lu_env *env, struct lu_object *o)
130 struct mdd_device *d = lu2mdd_dev(o->lo_dev);
131 struct lu_object *below;
132 struct lu_device *under;
135 under = &d->mdd_child->dd_lu_dev;
136 below = under->ld_ops->ldo_object_alloc(env, o->lo_header, under);
137 mdd_pdlock_init(lu2mdd_obj(o));
141 lu_object_add(o, below);
145 static int mdd_object_start(const struct lu_env *env, struct lu_object *o)
147 if (lu_object_exists(o))
148 return mdd_get_flags(env, lu2mdd_obj(o));
153 static void mdd_object_free(const struct lu_env *env, struct lu_object *o)
155 struct mdd_object *mdd = lu2mdd_obj(o);
161 static int mdd_object_print(const struct lu_env *env, void *cookie,
162 lu_printer_t p, const struct lu_object *o)
164 return (*p)(env, cookie, LUSTRE_MDD_NAME"-object@%p", o);
167 /* orphan handling is here */
168 static void mdd_object_delete(const struct lu_env *env,
171 struct mdd_object *mdd_obj = lu2mdd_obj(o);
172 struct thandle *handle = NULL;
175 if (lu2mdd_dev(o->lo_dev)->mdd_orphans == NULL)
178 if (mdd_obj->mod_flags & ORPHAN_OBJ) {
179 mdd_txn_param_build(env, lu2mdd_dev(o->lo_dev),
180 MDD_TXN_INDEX_DELETE_OP);
181 handle = mdd_trans_start(env, lu2mdd_dev(o->lo_dev));
183 CERROR("Cannot get thandle\n");
185 mdd_write_lock(env, mdd_obj);
186 /* let's remove obj from the orphan list */
187 __mdd_orphan_del(env, mdd_obj, handle);
188 mdd_write_unlock(env, mdd_obj);
189 mdd_trans_stop(env, lu2mdd_dev(o->lo_dev),
195 static struct lu_object_operations mdd_lu_obj_ops = {
196 .loo_object_init = mdd_object_init,
197 .loo_object_start = mdd_object_start,
198 .loo_object_free = mdd_object_free,
199 .loo_object_print = mdd_object_print,
200 .loo_object_delete = mdd_object_delete
203 struct mdd_object *mdd_object_find(const struct lu_env *env,
204 struct mdd_device *d,
205 const struct lu_fid *f)
207 struct lu_object *o, *lo;
208 struct mdd_object *m;
211 o = lu_object_find(env, mdd2lu_dev(d)->ld_site, f);
213 m = (struct mdd_object *)o;
215 lo = lu_object_locate(o->lo_header, mdd2lu_dev(d)->ld_type);
216 /* remote object can't be located and should be put then */
218 lu_object_put(env, o);
224 int mdd_get_flags(const struct lu_env *env, struct mdd_object *obj)
226 struct lu_attr *la = &mdd_env_info(env)->mti_la;
230 mdd_read_lock(env, obj);
231 rc = mdd_la_get(env, obj, la, BYPASS_CAPA);
232 mdd_read_unlock(env, obj);
234 mdd_flags_xlate(obj, la->la_flags);
238 void mdd_ref_add_internal(const struct lu_env *env, struct mdd_object *obj,
239 struct thandle *handle)
241 struct dt_object *next;
243 LASSERT(lu_object_exists(mdd2lu_obj(obj)));
244 next = mdd_object_child(obj);
245 next->do_ops->do_ref_add(env, next, handle);
248 void mdd_ref_del_internal(const struct lu_env *env, struct mdd_object *obj,
249 struct thandle *handle)
251 struct dt_object *next = mdd_object_child(obj);
254 LASSERT(lu_object_exists(mdd2lu_obj(obj)));
256 next->do_ops->do_ref_del(env, next, handle);
260 /* get only inode attributes */
261 int mdd_iattr_get(const struct lu_env *env, struct mdd_object *mdd_obj,
267 if (ma->ma_valid & MA_INODE)
270 rc = mdd_la_get(env, mdd_obj, &ma->ma_attr,
271 mdd_object_capa(env, mdd_obj));
273 ma->ma_valid |= MA_INODE;
277 /* get lov EA only */
278 static int __mdd_lmm_get(const struct lu_env *env,
279 struct mdd_object *mdd_obj, struct md_attr *ma)
284 if (ma->ma_valid & MA_LOV)
287 lmm_size = ma->ma_lmm_size;
288 rc = mdd_get_md(env, mdd_obj, ma->ma_lmm, &lmm_size,
291 ma->ma_valid |= MA_LOV;
292 ma->ma_lmm_size = lmm_size;
298 int mdd_lmm_get_locked(const struct lu_env *env, struct mdd_object *mdd_obj,
304 mdd_read_lock(env, mdd_obj);
305 rc = __mdd_lmm_get(env, mdd_obj, ma);
306 mdd_read_unlock(env, mdd_obj);
311 static int __mdd_lmv_get(const struct lu_env *env,
312 struct mdd_object *mdd_obj, struct md_attr *ma)
316 if (ma->ma_valid & MA_LMV)
319 rc = mdd_get_md(env, mdd_obj, ma->ma_lmv, &ma->ma_lmv_size,
322 ma->ma_valid |= MA_LMV;
328 static int mdd_attr_get_internal(const struct lu_env *env,
329 struct mdd_object *mdd_obj,
335 if (ma->ma_need & MA_INODE)
336 rc = mdd_iattr_get(env, mdd_obj, ma);
338 if (rc == 0 && ma->ma_need & MA_LOV) {
339 if (S_ISREG(mdd_object_type(mdd_obj)) ||
340 S_ISDIR(mdd_object_type(mdd_obj)))
341 rc = __mdd_lmm_get(env, mdd_obj, ma);
343 if (rc == 0 && ma->ma_need & MA_LMV) {
344 if (S_ISDIR(mdd_object_type(mdd_obj)))
345 rc = __mdd_lmv_get(env, mdd_obj, ma);
347 #ifdef CONFIG_FS_POSIX_ACL
348 if (rc == 0 && ma->ma_need & MA_ACL_DEF) {
349 if (S_ISDIR(mdd_object_type(mdd_obj)))
350 rc = mdd_acl_def_get(env, mdd_obj, ma);
353 CDEBUG(D_INODE, "after getattr rc = %d, ma_valid = "LPX64"\n",
358 int mdd_attr_get_internal_locked(const struct lu_env *env,
359 struct mdd_object *mdd_obj, struct md_attr *ma)
362 mdd_read_lock(env, mdd_obj);
363 rc = mdd_attr_get_internal(env, mdd_obj, ma);
364 mdd_read_unlock(env, mdd_obj);
369 * No permission check is needed.
371 static int mdd_attr_get(const struct lu_env *env, struct md_object *obj,
374 struct mdd_object *mdd_obj = md2mdd_obj(obj);
378 rc = mdd_attr_get_internal_locked(env, mdd_obj, ma);
383 * No permission check is needed.
385 static int mdd_xattr_get(const struct lu_env *env,
386 struct md_object *obj, struct lu_buf *buf,
389 struct mdd_object *mdd_obj = md2mdd_obj(obj);
390 struct dt_object *next;
395 LASSERT(lu_object_exists(&obj->mo_lu));
397 next = mdd_object_child(mdd_obj);
398 mdd_read_lock(env, mdd_obj);
399 rc = next->do_ops->do_xattr_get(env, next, buf, name,
400 mdd_object_capa(env, mdd_obj));
401 mdd_read_unlock(env, mdd_obj);
407 * Permission check is done when open,
408 * no need check again.
410 static int mdd_readlink(const struct lu_env *env, struct md_object *obj,
413 struct mdd_object *mdd_obj = md2mdd_obj(obj);
414 struct dt_object *next;
419 LASSERT(lu_object_exists(&obj->mo_lu));
421 next = mdd_object_child(mdd_obj);
422 mdd_read_lock(env, mdd_obj);
423 rc = next->do_body_ops->dbo_read(env, next, buf, &pos,
424 mdd_object_capa(env, mdd_obj));
425 mdd_read_unlock(env, mdd_obj);
429 static int mdd_xattr_list(const struct lu_env *env, struct md_object *obj,
432 struct mdd_object *mdd_obj = md2mdd_obj(obj);
433 struct dt_object *next;
438 LASSERT(lu_object_exists(&obj->mo_lu));
440 next = mdd_object_child(mdd_obj);
441 mdd_read_lock(env, mdd_obj);
442 rc = next->do_ops->do_xattr_list(env, next, buf,
443 mdd_object_capa(env, mdd_obj));
444 mdd_read_unlock(env, mdd_obj);
449 int mdd_object_create_internal(const struct lu_env *env,
450 struct mdd_object *obj, struct md_attr *ma,
451 struct thandle *handle)
453 struct dt_object *next;
454 struct lu_attr *attr = &ma->ma_attr;
458 if (!lu_object_exists(mdd2lu_obj(obj))) {
459 next = mdd_object_child(obj);
460 rc = next->do_ops->do_create(env, next, attr, handle);
464 LASSERT(ergo(rc == 0, lu_object_exists(mdd2lu_obj(obj))));
470 int mdd_attr_set_internal(const struct lu_env *env, struct mdd_object *o,
471 const struct lu_attr *attr, struct thandle *handle,
474 struct dt_object *next;
477 LASSERT(lu_object_exists(mdd2lu_obj(o)));
478 next = mdd_object_child(o);
479 rc = next->do_ops->do_attr_set(env, next, attr, handle,
480 mdd_object_capa(env, o));
481 #ifdef CONFIG_FS_POSIX_ACL
482 if (!rc && (attr->la_valid & LA_MODE) && needacl)
483 rc = mdd_acl_chmod(env, o, attr->la_mode, handle);
488 int mdd_attr_set_internal_locked(const struct lu_env *env,
489 struct mdd_object *o,
490 const struct lu_attr *attr,
491 struct thandle *handle, int needacl)
495 needacl = needacl && (attr->la_valid & LA_MODE);
498 mdd_write_lock(env, o);
500 rc = mdd_attr_set_internal(env, o, attr, handle, needacl);
503 mdd_write_unlock(env, o);
507 static int __mdd_xattr_set(const struct lu_env *env, struct mdd_object *o,
508 const struct lu_buf *buf, const char *name,
509 int fl, struct thandle *handle)
511 struct dt_object *next;
512 struct lustre_capa *capa = mdd_object_capa(env, o);
516 LASSERT(lu_object_exists(mdd2lu_obj(o)));
517 next = mdd_object_child(o);
518 if (buf->lb_buf && buf->lb_len > 0) {
519 rc = next->do_ops->do_xattr_set(env, next, buf, name, 0, handle,
521 } else if (buf->lb_buf == NULL && buf->lb_len == 0) {
522 rc = next->do_ops->do_xattr_del(env, next, name, handle, capa);
527 /* this gives the same functionality as the code between
528 * sys_chmod and inode_setattr
529 * chown_common and inode_setattr
530 * utimes and inode_setattr
531 * This API is ported from mds_fix_attr but remove some unnecesssary stuff.
534 int mdd_fix_attr(const struct lu_env *env, struct mdd_object *obj,
537 struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la;
538 struct md_ucred *uc = md_ucred(env);
539 time_t now = CURRENT_SECONDS;
546 /* Do not permit change file type */
547 if (la->la_valid & LA_TYPE)
550 /* They should not be processed by setattr */
551 if (la->la_valid & (LA_NLINK | LA_RDEV | LA_BLKSIZE))
554 rc = mdd_la_get(env, obj, tmp_la, BYPASS_CAPA);
558 if (mdd_is_immutable(obj) || mdd_is_append(obj)) {
561 * If only change flags of the object, we should
562 * let it pass, but also need capability check
563 * here if (!capable(CAP_LINUX_IMMUTABLE)),
564 * fix it, when implement capable in mds
566 if (la->la_valid & ~LA_FLAGS)
569 if (!mdd_capable(uc, CAP_LINUX_IMMUTABLE))
572 if ((uc->mu_fsuid != tmp_la->la_uid) &&
573 !mdd_capable(uc, CAP_FOWNER))
577 * According to Ext3 implementation on this, the
578 * Ctime will be changed, but not clear why?
581 la->la_valid |= LA_CTIME;
585 /* Check for setting the obj time. */
586 if ((la->la_valid & (LA_MTIME | LA_ATIME | LA_CTIME)) &&
587 !(la->la_valid & ~(LA_MTIME | LA_ATIME | LA_CTIME))) {
588 rc = __mdd_permission_internal(env, obj, MAY_WRITE, 0);
593 /* Make sure a caller can chmod. */
594 if (la->la_valid & LA_MODE) {
596 * Bypass la_vaild == LA_MODE,
597 * this is for changing file with SUID or SGID.
599 if ((la->la_valid & ~LA_MODE) &&
600 (uc->mu_fsuid != tmp_la->la_uid) &&
601 !mdd_capable(uc, CAP_FOWNER))
604 if (la->la_mode == (umode_t) -1)
605 la->la_mode = tmp_la->la_mode;
607 la->la_mode = (la->la_mode & S_IALLUGO) |
608 (tmp_la->la_mode & ~S_IALLUGO);
610 /* Also check the setgid bit! */
611 if (!mdd_in_group_p(uc, (la->la_valid & LA_GID) ? la->la_gid :
612 tmp_la->la_gid) && !mdd_capable(uc, CAP_FSETID))
613 la->la_mode &= ~S_ISGID;
615 la->la_mode = tmp_la->la_mode;
618 /* Make sure a caller can chown. */
619 if (la->la_valid & LA_UID) {
620 if (la->la_uid == (uid_t) -1)
621 la->la_uid = tmp_la->la_uid;
622 if (((uc->mu_fsuid != tmp_la->la_uid) ||
623 (la->la_uid != tmp_la->la_uid)) &&
624 !mdd_capable(uc, CAP_CHOWN))
628 * If the user or group of a non-directory has been
629 * changed by a non-root user, remove the setuid bit.
630 * 19981026 David C Niemi <niemi@tux.org>
632 * Changed this to apply to all users, including root,
633 * to avoid some races. This is the behavior we had in
634 * 2.0. The check for non-root was definitely wrong
635 * for 2.2 anyway, as it should have been using
636 * CAP_FSETID rather than fsuid -- 19990830 SD.
638 if (((tmp_la->la_mode & S_ISUID) == S_ISUID) &&
639 !S_ISDIR(tmp_la->la_mode)) {
640 la->la_mode &= ~S_ISUID;
641 la->la_valid |= LA_MODE;
645 /* Make sure caller can chgrp. */
646 if (la->la_valid & LA_GID) {
647 if (la->la_gid == (gid_t) -1)
648 la->la_gid = tmp_la->la_gid;
649 if (((uc->mu_fsuid != tmp_la->la_uid) ||
650 ((la->la_gid != tmp_la->la_gid) &&
651 !mdd_in_group_p(uc, la->la_gid))) &&
652 !mdd_capable(uc, CAP_CHOWN))
656 * Likewise, if the user or group of a non-directory
657 * has been changed by a non-root user, remove the
658 * setgid bit UNLESS there is no group execute bit
659 * (this would be a file marked for mandatory
660 * locking). 19981026 David C Niemi <niemi@tux.org>
662 * Removed the fsuid check (see the comment above) --
665 if (((tmp_la->la_mode & (S_ISGID | S_IXGRP)) ==
666 (S_ISGID | S_IXGRP)) && !S_ISDIR(tmp_la->la_mode)) {
667 la->la_mode &= ~S_ISGID;
668 la->la_valid |= LA_MODE;
672 /* For tuncate (or setsize), we should have MAY_WRITE perm */
673 if (la->la_valid & (LA_SIZE | LA_BLOCKS)) {
674 rc = __mdd_permission_internal(env, obj, MAY_WRITE, 0);
679 * For the "Size-on-MDS" setattr update, merge coming
680 * attributes with the set in the inode. BUG 10641
682 if ((la->la_valid & LA_ATIME) &&
683 (la->la_atime < tmp_la->la_atime))
684 la->la_valid &= ~LA_ATIME;
686 if ((la->la_valid & LA_CTIME) &&
687 (la->la_ctime < tmp_la->la_ctime))
688 la->la_valid &= ~(LA_MTIME | LA_CTIME);
690 if (!(la->la_valid & LA_MTIME) && (now > tmp_la->la_mtime)) {
692 la->la_valid |= LA_MTIME;
696 /* For last, ctime must be fixed */
697 if (!(la->la_valid & LA_CTIME) && (now > tmp_la->la_ctime)) {
699 la->la_valid |= LA_CTIME;
705 /* set attr and LOV EA at once, return updated attr */
706 static int mdd_attr_set(const struct lu_env *env, struct md_object *obj,
707 const struct md_attr *ma)
709 struct mdd_object *mdd_obj = md2mdd_obj(obj);
710 struct mdd_device *mdd = mdo2mdd(obj);
711 struct thandle *handle;
712 struct lov_mds_md *lmm = NULL;
713 int rc, lmm_size = 0, max_size = 0;
714 struct lu_attr *la_copy = &mdd_env_info(env)->mti_la_for_fix;
717 mdd_txn_param_build(env, mdd, MDD_TXN_ATTR_SET_OP);
718 handle = mdd_trans_start(env, mdd);
720 RETURN(PTR_ERR(handle));
721 /*TODO: add lock here*/
722 /* start a log jounal handle if needed */
723 if (S_ISREG(mdd_object_type(mdd_obj)) &&
724 ma->ma_attr.la_valid & (LA_UID | LA_GID)) {
725 max_size = mdd_lov_mdsize(env, mdd);
726 OBD_ALLOC(lmm, max_size);
729 GOTO(cleanup, rc = -ENOMEM);
731 rc = mdd_get_md_locked(env, mdd_obj, lmm, &lmm_size,
738 if (ma->ma_attr.la_valid & (ATTR_MTIME | ATTR_CTIME))
739 CDEBUG(D_INODE, "setting mtime "LPU64", ctime "LPU64"\n",
740 ma->ma_attr.la_mtime, ma->ma_attr.la_ctime);
742 *la_copy = ma->ma_attr;
743 mdd_read_lock(env, mdd_obj);
744 rc = mdd_fix_attr(env, mdd_obj, la_copy);
745 mdd_read_unlock(env, mdd_obj);
749 if (la_copy->la_valid & LA_FLAGS) {
750 rc = mdd_attr_set_internal_locked(env, mdd_obj, la_copy,
753 mdd_flags_xlate(mdd_obj, la_copy->la_flags);
754 } else if (la_copy->la_valid) { /* setattr */
755 rc = mdd_attr_set_internal_locked(env, mdd_obj, la_copy,
757 /* journal chown/chgrp in llog, just like unlink */
758 if (rc == 0 && lmm_size){
759 /*TODO set_attr llog */
763 if (rc == 0 && ma->ma_valid & MA_LOV) {
766 mode = mdd_object_type(mdd_obj);
767 if (S_ISREG(mode) || S_ISDIR(mode)) {
768 /*TODO check permission*/
769 rc = mdd_lov_set_md(env, NULL, mdd_obj, ma->ma_lmm,
770 ma->ma_lmm_size, handle, 1);
775 mdd_trans_stop(env, mdd, rc, handle);
776 if (rc == 0 && (lmm != NULL && lmm_size > 0 )) {
777 /*set obd attr, if needed*/
778 rc = mdd_lov_setattr_async(env, mdd_obj, lmm, lmm_size);
781 OBD_FREE(lmm, max_size);
787 int mdd_xattr_set_txn(const struct lu_env *env, struct mdd_object *obj,
788 const struct lu_buf *buf, const char *name, int fl,
789 struct thandle *handle)
794 mdd_write_lock(env, obj);
795 rc = __mdd_xattr_set(env, obj, buf, name, fl, handle);
796 mdd_write_unlock(env, obj);
801 static int mdd_xattr_sanity_check(const struct lu_env *env,
802 struct mdd_object *obj)
804 struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la;
805 struct md_ucred *uc = md_ucred(env);
809 if (mdd_is_immutable(obj) || mdd_is_append(obj))
812 mdd_read_lock(env, obj);
813 rc = mdd_la_get(env, obj, tmp_la, BYPASS_CAPA);
814 mdd_read_unlock(env, obj);
818 if ((uc->mu_fsuid != tmp_la->la_uid) && !mdd_capable(uc, CAP_FOWNER))
824 static int mdd_xattr_set(const struct lu_env *env, struct md_object *obj,
825 const struct lu_buf *buf, const char *name, int fl)
827 struct lu_attr *la_copy = &mdd_env_info(env)->mti_la_for_fix;
828 struct mdd_object *mdd_obj = md2mdd_obj(obj);
829 struct mdd_device *mdd = mdo2mdd(obj);
830 struct thandle *handle;
834 rc = mdd_xattr_sanity_check(env, mdd_obj);
838 mdd_txn_param_build(env, mdd, MDD_TXN_XATTR_SET_OP);
839 handle = mdd_trans_start(env, mdd);
841 RETURN(PTR_ERR(handle));
843 rc = mdd_xattr_set_txn(env, md2mdd_obj(obj), buf, name,
846 la_copy->la_ctime = CURRENT_SECONDS;
847 la_copy->la_valid = LA_CTIME;
848 rc = mdd_attr_set_internal_locked(env, mdd_obj, la_copy, handle, 0);
850 mdd_trans_stop(env, mdd, rc, handle);
855 static int __mdd_xattr_del(const struct lu_env *env,struct mdd_device *mdd,
856 struct mdd_object *obj,
857 const char *name, struct thandle *handle)
859 struct dt_object *next;
861 LASSERT(lu_object_exists(mdd2lu_obj(obj)));
862 next = mdd_object_child(obj);
863 return next->do_ops->do_xattr_del(env, next, name, handle,
864 mdd_object_capa(env, obj));
867 int mdd_xattr_del(const struct lu_env *env, struct md_object *obj,
870 struct lu_attr *la_copy = &mdd_env_info(env)->mti_la_for_fix;
871 struct mdd_object *mdd_obj = md2mdd_obj(obj);
872 struct mdd_device *mdd = mdo2mdd(obj);
873 struct thandle *handle;
877 rc = mdd_xattr_sanity_check(env, mdd_obj);
881 mdd_txn_param_build(env, mdd, MDD_TXN_XATTR_SET_OP);
882 handle = mdd_trans_start(env, mdd);
884 RETURN(PTR_ERR(handle));
886 mdd_write_lock(env, mdd_obj);
887 rc = __mdd_xattr_del(env, mdd, md2mdd_obj(obj), name, handle);
888 mdd_write_unlock(env, mdd_obj);
890 la_copy->la_ctime = CURRENT_SECONDS;
891 la_copy->la_valid = LA_CTIME;
892 rc = mdd_attr_set_internal(env, mdd_obj, la_copy, handle, 0);
895 mdd_trans_stop(env, mdd, rc, handle);
901 static int mdd_ref_del(const struct lu_env *env, struct md_object *obj,
904 struct lu_attr *la_copy = &mdd_env_info(env)->mti_la_for_fix;
905 struct mdd_object *mdd_obj = md2mdd_obj(obj);
906 struct mdd_device *mdd = mdo2mdd(obj);
907 struct thandle *handle;
911 rc = mdd_log_txn_param_build(env, obj, ma, MDD_TXN_UNLINK_OP);
915 handle = mdd_trans_start(env, mdd);
919 mdd_write_lock(env, mdd_obj);
921 rc = mdd_unlink_sanity_check(env, NULL, mdd_obj, ma);
925 mdd_ref_del_internal(env, mdd_obj, handle);
927 if (S_ISDIR(lu_object_attr(&obj->mo_lu))) {
929 mdd_ref_del_internal(env, mdd_obj, handle);
932 la_copy->la_ctime = CURRENT_SECONDS;
933 la_copy->la_valid = LA_CTIME;
934 rc = mdd_attr_set_internal(env, mdd_obj, la_copy, handle, 0);
938 rc = mdd_finish_unlink(env, mdd_obj, ma, handle);
942 mdd_write_unlock(env, mdd_obj);
943 mdd_trans_stop(env, mdd, rc, handle);
947 /* partial operation */
948 static int mdd_oc_sanity_check(const struct lu_env *env,
949 struct mdd_object *obj,
955 switch (ma->ma_attr.la_mode & S_IFMT) {
972 static int mdd_object_create(const struct lu_env *env,
973 struct md_object *obj,
974 const struct md_create_spec *spec,
978 struct mdd_device *mdd = mdo2mdd(obj);
979 struct mdd_object *mdd_obj = md2mdd_obj(obj);
980 const struct lu_fid *pfid = spec->u.sp_pfid;
981 struct thandle *handle;
985 mdd_txn_param_build(env, mdd, MDD_TXN_OBJECT_CREATE_OP);
986 handle = mdd_trans_start(env, mdd);
988 RETURN(PTR_ERR(handle));
990 mdd_write_lock(env, mdd_obj);
991 rc = mdd_oc_sanity_check(env, mdd_obj, ma);
995 rc = mdd_object_create_internal(env, mdd_obj, ma, handle);
999 if (spec->sp_cr_flags & MDS_CREATE_SLAVE_OBJ) {
1000 /* If creating the slave object, set slave EA here. */
1001 int lmv_size = spec->u.sp_ea.eadatalen;
1002 struct lmv_stripe_md *lmv;
1004 lmv = (struct lmv_stripe_md *)spec->u.sp_ea.eadata;
1005 LASSERT(lmv != NULL && lmv_size > 0);
1007 rc = __mdd_xattr_set(env, mdd_obj,
1008 mdd_buf_get_const(env, lmv, lmv_size),
1009 MDS_LMV_MD_NAME, 0, handle);
1012 pfid = spec->u.sp_ea.fid;
1014 CDEBUG(D_INFO, "Set slave ea "DFID", eadatalen %d, rc %d\n",
1015 PFID(mdo2fid(mdd_obj)), spec->u.sp_ea.eadatalen, rc);
1017 rc = mdd_attr_set_internal(env, mdd_obj, &ma->ma_attr, handle, 0);
1019 #ifdef CONFIG_FS_POSIX_ACL
1020 if (spec->sp_cr_flags & MDS_CREATE_RMT_ACL) {
1021 struct lu_buf *buf = &mdd_env_info(env)->mti_buf;
1023 buf->lb_buf = (void *)spec->u.sp_ea.eadata;
1024 buf->lb_len = spec->u.sp_ea.eadatalen;
1025 if ((buf->lb_len > 0) && (buf->lb_buf != NULL)) {
1026 rc = __mdd_acl_init(env, mdd_obj, buf,
1027 &ma->ma_attr.la_mode,
1032 ma->ma_attr.la_valid |= LA_MODE;
1036 rc = mdd_object_initialize(env, pfid, mdd_obj, ma, handle);
1040 mdd_write_unlock(env, mdd_obj);
1042 rc = mdd_attr_get_internal_locked(env, mdd_obj, ma);
1044 mdd_trans_stop(env, mdd, rc, handle);
1049 * XXX: Do we need permission checks here?
1051 static int mdd_ref_add(const struct lu_env *env,
1052 struct md_object *obj)
1054 struct lu_attr *la_copy = &mdd_env_info(env)->mti_la_for_fix;
1055 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1056 struct mdd_device *mdd = mdo2mdd(obj);
1057 struct thandle *handle;
1061 mdd_txn_param_build(env, mdd, MDD_TXN_XATTR_SET_OP);
1062 handle = mdd_trans_start(env, mdd);
1066 mdd_write_lock(env, mdd_obj);
1067 rc = mdd_link_sanity_check(env, NULL, mdd_obj);
1069 mdd_ref_add_internal(env, mdd_obj, handle);
1070 mdd_write_unlock(env, mdd_obj);
1072 la_copy->la_ctime = CURRENT_SECONDS;
1073 la_copy->la_valid = LA_CTIME;
1074 rc = mdd_attr_set_internal(env, mdd_obj, la_copy, handle, 0);
1076 mdd_trans_stop(env, mdd, 0, handle);
1081 /* do NOT or the MAY_*'s, you'll get the weakest */
1082 static int accmode(struct mdd_object *mdd_obj, int flags)
1087 /* Sadly, NFSD reopens a file repeatedly during operation, so the
1088 * "acc_mode = 0" allowance for newly-created files isn't honoured.
1089 * NFSD uses the MDS_OPEN_OWNEROVERRIDE flag to say that a file
1090 * owner can write to a file even if it is marked readonly to hide
1091 * its brokenness. (bug 5781) */
1092 if (flags & MDS_OPEN_OWNEROVERRIDE && inode->i_uid == current->fsuid)
1095 if (flags & FMODE_READ)
1097 if (flags & (FMODE_WRITE | MDS_OPEN_TRUNC | MDS_OPEN_APPEND))
1099 if (flags & MDS_FMODE_EXEC)
1104 static int mdd_open_sanity_check(const struct lu_env *env,
1105 struct mdd_object *obj, int flag)
1107 struct lu_attr *tmp_la = &mdd_env_info(env)->mti_la;
1108 int mode = accmode(obj, flag);
1113 if (mdd_is_dead_obj(obj))
1116 rc = mdd_la_get(env, obj, tmp_la, BYPASS_CAPA);
1120 if (S_ISLNK(tmp_la->la_mode))
1123 if (S_ISDIR(tmp_la->la_mode) && (mode & MAY_WRITE))
1126 if (!(flag & MDS_OPEN_CREATED)) {
1127 rc = __mdd_permission_internal(env, obj, mode, 0);
1132 if (S_ISFIFO(tmp_la->la_mode) || S_ISSOCK(tmp_la->la_mode) ||
1133 S_ISBLK(tmp_la->la_mode) || S_ISCHR(tmp_la->la_mode))
1134 flag &= ~MDS_OPEN_TRUNC;
1137 * For writing append-only file must open it with append mode.
1139 if (mdd_is_append(obj)) {
1140 if ((flag & FMODE_WRITE) && !(flag & MDS_OPEN_APPEND))
1142 if (flag & MDS_OPEN_TRUNC)
1148 * Now, flag -- O_NOATIME does not be packed by client.
1150 if (flag & O_NOATIME) {
1151 struct md_ucred *uc = md_ucred(env);
1153 if (uc->mu_fsuid != tmp_la->la_uid &&
1154 !mdd_capable(uc, CAP_FOWNER))
1162 static int mdd_open(const struct lu_env *env, struct md_object *obj,
1165 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1168 mdd_write_lock(env, mdd_obj);
1170 rc = mdd_open_sanity_check(env, mdd_obj, flags);
1172 mdd_obj->mod_count++;
1174 mdd_write_unlock(env, mdd_obj);
1178 /* return md_attr back,
1179 * if it is last unlink then return lov ea + llog cookie*/
1180 int mdd_object_kill(const struct lu_env *env, struct mdd_object *obj,
1186 mdd_set_dead_obj(obj);
1187 if (S_ISREG(mdd_object_type(obj))) {
1188 /* Return LOV & COOKIES unconditionally here. We clean evth up.
1189 * Caller must be ready for that. */
1190 rc = __mdd_lmm_get(env, obj, ma);
1191 if ((ma->ma_valid & MA_LOV))
1192 rc = mdd_unlink_log(env, mdo2mdd(&obj->mod_obj),
1199 * No permission check is needed.
1201 static int mdd_close(const struct lu_env *env, struct md_object *obj,
1205 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1208 mdd_write_lock(env, mdd_obj);
1209 /* release open count */
1210 mdd_obj->mod_count --;
1212 rc = mdd_iattr_get(env, mdd_obj, ma);
1213 if (rc == 0 && mdd_obj->mod_count == 0) {
1214 if (ma->ma_attr.la_nlink == 0)
1215 rc = mdd_object_kill(env, mdd_obj, ma);
1217 mdd_write_unlock(env, mdd_obj);
1222 * Permission check is done when open,
1223 * no need check again.
1225 static int mdd_readpage_sanity_check(const struct lu_env *env,
1226 struct mdd_object *obj)
1228 struct dt_object *next = mdd_object_child(obj);
1232 if (S_ISDIR(mdd_object_type(obj)) && dt_try_as_dir(env, next))
1234 rc = mdd_permission_internal(env, obj, MAY_READ);
1244 static int mdd_dir_page_build(const struct lu_env *env, int first,
1245 void *area, int nob, struct dt_it_ops *iops,
1246 struct dt_it *it, __u32 *start, __u32 *end,
1247 struct lu_dirent **last)
1249 struct lu_fid *fid2 = &mdd_env_info(env)->mti_fid2;
1250 struct mdd_thread_info *info = mdd_env_info(env);
1251 struct lu_fid *fid = &info->mti_fid;
1253 struct lu_dirent *ent;
1256 memset(area, 0, sizeof (struct lu_dirpage));
1257 area += sizeof (struct lu_dirpage);
1258 nob -= sizeof (struct lu_dirpage);
1261 LASSERT(nob > sizeof *ent);
1271 name = (char *)iops->key(env, it);
1272 len = iops->key_size(env, it);
1274 fid = (struct lu_fid *)iops->rec(env, it);
1275 fid_be_to_cpu(fid2, fid);
1277 recsize = (sizeof(*ent) + len + 3) & ~3;
1278 hash = iops->store(env, it);
1281 CDEBUG(D_INFO, "%p %p %d "DFID": %#8.8x (%d) \"%*.*s\"\n",
1282 name, ent, nob, PFID(fid2), hash, len, len, len, name);
1284 if (nob >= recsize) {
1285 fid_be_to_cpu(&ent->lde_fid, fid);
1286 fid_cpu_to_le(&ent->lde_fid, &ent->lde_fid);
1287 ent->lde_hash = hash;
1288 ent->lde_namelen = cpu_to_le16(len);
1289 ent->lde_reclen = cpu_to_le16(recsize);
1290 memcpy(ent->lde_name, name, len);
1291 if (first && ent == area)
1294 ent = (void *)ent + recsize;
1296 result = iops->next(env, it);
1299 * record doesn't fit into page, enlarge previous one.
1301 LASSERT(*last != NULL);
1302 (*last)->lde_reclen =
1303 cpu_to_le16(le16_to_cpu((*last)->lde_reclen) +
1307 } while (result == 0);
1312 static int __mdd_readpage(const struct lu_env *env, struct mdd_object *obj,
1313 const struct lu_rdpg *rdpg)
1316 struct dt_object *next = mdd_object_child(obj);
1317 struct dt_it_ops *iops;
1319 struct lu_dirent *last;
1326 LASSERT(rdpg->rp_pages != NULL);
1327 LASSERT(next->do_index_ops != NULL);
1329 if (rdpg->rp_count <= 0)
1333 * iterate through directory and fill pages from @rdpg
1335 iops = &next->do_index_ops->dio_it;
1336 it = iops->init(env, next, 0, mdd_object_capa(env, obj));
1340 rc = iops->load(env, it, rdpg->rp_hash);
1344 * Iterator didn't find record with exactly the key requested.
1346 * It is currently either
1348 * - positioned above record with key less than
1349 * requested---skip it.
1351 * - or not positioned at all (is in IAM_IT_SKEWED
1352 * state)---position it on the next item.
1354 rc = iops->next(env, it);
1359 * At this point and across for-loop:
1361 * rc == 0 -> ok, proceed.
1362 * rc > 0 -> end of directory.
1365 for (i = 0, nob = rdpg->rp_count; rc == 0 && nob > 0;
1366 i++, nob -= CFS_PAGE_SIZE) {
1367 LASSERT(i < rdpg->rp_npages);
1368 pg = rdpg->rp_pages[i];
1369 rc = mdd_dir_page_build(env, !i, kmap(pg),
1370 min_t(int, nob, CFS_PAGE_SIZE), iops,
1371 it, &hash_start, &hash_end, &last);
1372 if (rc != 0 || i == rdpg->rp_npages - 1)
1373 last->lde_reclen = 0;
1384 struct lu_dirpage *dp;
1386 dp = kmap(rdpg->rp_pages[0]);
1387 dp->ldp_hash_start = rdpg->rp_hash;
1388 dp->ldp_hash_end = hash_end;
1391 * No pages were processed, mark this.
1393 dp->ldp_flags |= LDF_EMPTY;
1394 dp->ldp_flags = cpu_to_le16(dp->ldp_flags);
1395 kunmap(rdpg->rp_pages[0]);
1398 iops->fini(env, it);
1403 static int mdd_readpage(const struct lu_env *env, struct md_object *obj,
1404 const struct lu_rdpg *rdpg)
1406 struct dt_object *next;
1407 struct mdd_object *mdd_obj = md2mdd_obj(obj);
1411 LASSERT(lu_object_exists(mdd2lu_obj(mdd_obj)));
1412 next = mdd_object_child(mdd_obj);
1414 mdd_read_lock(env, mdd_obj);
1415 rc = mdd_readpage_sanity_check(env, mdd_obj);
1417 GOTO(out_unlock, rc);
1419 rc = __mdd_readpage(env, mdd_obj, rdpg);
1423 mdd_read_unlock(env, mdd_obj);
1427 struct md_object_operations mdd_obj_ops = {
1428 .moo_permission = mdd_permission,
1429 .moo_attr_get = mdd_attr_get,
1430 .moo_attr_set = mdd_attr_set,
1431 .moo_xattr_get = mdd_xattr_get,
1432 .moo_xattr_set = mdd_xattr_set,
1433 .moo_xattr_list = mdd_xattr_list,
1434 .moo_xattr_del = mdd_xattr_del,
1435 .moo_object_create = mdd_object_create,
1436 .moo_ref_add = mdd_ref_add,
1437 .moo_ref_del = mdd_ref_del,
1438 .moo_open = mdd_open,
1439 .moo_close = mdd_close,
1440 .moo_readpage = mdd_readpage,
1441 .moo_readlink = mdd_readlink,
1442 .moo_capa_get = mdd_capa_get